Scheduled tasks: allow setting model and effort when creating/updating a routine

Problem

The scheduled-task tools (create_scheduled_task / update_scheduled_task) have no model or effort parameter, and the task config does not persist one. Every run inherits whatever model the app's model selector has at the moment the task fires.

What happened

• From a chat session I created 12 routines and asked Claude to run them all on Sonnet 5.5. Claude could not set the model: the tools have no field for it.
• Overnight the app's selector changed. 37 of the 50 runs executed on Opus 5.5 instead of Sonnet.
• That exhausted my usage limit (Max plan) and the remaining 24 runs failed with "You've hit your session limit".
• The only workaround today is opening each routine in the UI and setting the model by hand, one by one.

Request

1. Add optional model and effort parameters to create_scheduled_task and update_scheduled_task, persisted per task.
2. A routine with a model set runs on that model regardless of the app's selector.
3. Return the current model / effort in list_scheduled_tasks, so Claude can verify it.

Why it matters

Scheduled routines run unattended, often in parallel. The model is a cost and quality decision made when the routine is designed, not a UI state that can drift at run time. If Claude can create a routine from chat, it should be able to pin its model too.

Environment

Claude desktop app (Code tab), macOS, scheduled tasks created via the scheduled-tasks MCP tools.